研究目的
The goal of this work is to answer two main questions. The first one is to know if it is possible to identify machine tool’s behavioural groups (or clusters) using data from three of the sensors used in [6]. The second question is to discover the proper number of clusters and which statistical features could be used in order to use them to foresee the machine tool’s behaviour for the use in the context of predictive maintenance.
研究成果
This work presented the possibility to identify machine tool’s operational groups using data from sensors of the platform temperature, oxygen percentage in the process chamber and process chamber pressure. The outcomes are a first step towards the implementation of a condition monitoring system that enables SLM machine tools for predictive maintenance solutions, because the actual condition of the machine and process is monitored and an early fault detection of specific systems can be performed.
研究不足
The existing SLM machine tools are often not able to ensure the product quality, due to several factors, including failures occurred during manufacturing processes. Such machine tools are not equipped with analytics tools that evaluate the machine data, which makes the identification of the factors and the conditions of operation a challenging task.
1：Experimental Design and Method Selection
The methodology performed in this work includes raw machine tool sensor data acquisition, sensor data pre-processing, cluster analysis and assessment. The algorithm k-means was used to perform the identification of clusters.
2：Sample Selection and Data Sources
Data from 206 manufacturing processes were acquired and stored. The SLM machine model used to gather the data is the SLM 250HL from the company SLM Solutions AG, Germany.
3：List of Experimental Equipment and Materials
The data from three sensors related specifically to the machine operation were chosen and pre-processed. Those sensors are the platform temperature (T), oxygen percentage within the process chamber (O) and the process chamber pressure (P).
4：Experimental Procedures and Operational Workflow
The pre-processing stage was performed by collecting the 206 files using the software tool Spyder 3. Then, only data within the interval from the first layer until the end of the manufacturing process were considered. The data of the three sensors were separated into three independent columns according to the manufacturing process. After that, the statistical features minimum value, skewness, maximum value, mode, median, average, and standard deviation from each column were calculated and stored in a two-dimensional matrix.
5：Data Analysis Methods
The cluster analysis was carried out using the algorithm k-means. The number of clusters was defined using the elbow method. The results of this analysis were assessed in order to find the proper number of clusters according to the statistical features from those three sensors.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
